We thank the reviewer for positive and constructive feedback. Below, we will address the comments one by one.



Comment 1: I take the point that the oceanic DA with an elaborated EnKF scheme carries a lot of the interest to the authors. However, it seems that the combination NorESM+CMIP6 external forcing carries most of the skill in decadal predictions. As we know, this can be seen in other ESMs, too. And of course this usually is a good sign: if the combination ESM + external forcing delivers good results, so that assimilation does not have to repair too many "shortcomings". To provoke the authors, albeit in a friendly manner: should the prediction community rather invest in better models than in sophisticated assimilation?



The question raised is a hot issue, especially since recent findings suggest that external forcings have a stronger influence on extratropical atmospheric circulation variability than previously thought (e.g., Athanasiadis et al., 2020; Liguori et al., 2020; Drews et al., 2021; Klavans et al., 2021). We will add some short discussion on this to the revised manuscript based on the text below.



Improving the climate models utilized in the prediction systems is potentially very important for achieving more skilful near-term climate predictions (e.g., Athanasiadis et al., 2020). Increasing the sophistication of data assimilation has its limits with respect to mitigating effects from shortcomings in the dynamical behaviour of the models and errors in their response to external forcings. However, one can augment the state estimation with model parameters or external forcings and use advanced data assimilation to tune model parameters (Annan, 2005) and mitigate model bias, an approach that is currently being tested with our system. While skill gains from improving the models (Athanasiadis et al., 2020) may outweigh benefits from refining assimilation schemes, the combination is required to maximize the skill. Hence there is a need to both improve models and methods to assimilate observations. We have seen from weather and seasonal forecasting how improvements in both (as well as observations and computing power) have continued to lead to enhanced prediction skill (Bauer et al., 2015). Data assimilation methods have been mainly developed for numerical weather prediction and they need to be adapted for the climate system with multiscale interaction, an effort that will take some time. For NorCPM, the benefits from model and assimilation development are not necessarily independent, because improving the model will also improve our initialization capability as the ensemble covariance would improve (Counillon et al., 2021).



We argue that there is merit from the use of sophisticated assimilation and its further development. While the forced ocean–sea ice (FOSI) initialisation approach (that solely constrains surface fluxes) has been proven rather successful for multi-year climate prediction (Yeager et al., 2018), Polkova et al. (2019) found significant skill improvements from EnKF ocean assimilation in addition to constraining the atmosphere via nudging. The drift in prediction systems is another clear evidence that both model and assimilation improvements are needed. Progress has been made on reducing model biases but it is uncertain whether and when they can be completely eliminated (including conditional biases related to the forced trend). To date, the benefits from reducing initialisation shock and forecast drift or from having optimal spread in initial conditions are not well explored and elaborated data assimilation schemes are needed to appropriately deal with observational uncertainties and sparseness. How to best handle errors in the forced model trend during initialisation (e.g., Chikamoto et al., 2019) is a particular challenge that warrants further investigation. 



The reviewer highlights the predictive potential due to external forcings. Using the historical all-forcing experiment of the Multi-Model Large Ensemble Archive (Deser et al., 2020; Klavans et al., 2021; Liguori et al., 2020) as an additional benchmark could help assess how improved response to external forcings may impact prediction skill. There is growing evidence, however, that current generation climate models systematically underestimate the influence of SST variations and external forcing variability on extratropical atmospheric variability, particularly related to the North Atlantic Oscillation (e.g., Scaife and Smith, 2018). As a consequence, the amplitude of the forced climate signal (either from surface boundary conditions or external forcings) is underestimated relative to the intrinsic climate variability. While post-processing methods relying on large ensembles have been proposed to mitigate this shortcoming (Smith et al. 2020), improving this aspect in the next model generation should be a key priority for the prediction community. To this end, we are investigating key processes in NorCPM—like atmospheric wave breaking and weather regime shifts in relation to boundary conditions and also ocean dynamics—that have been identified as essential for skilful near-term climate prediction (Athanasiadis et al., 2020). Rather than moving resources from data assimilation to ESM development, we plan to use data assimilation increasingly to inform the ESM development to improve processes and dynamics key to seasonal-to-decadal climate prediction. In this manuscript we focused on data assimilation, but in future work we will assess improvements related to the model being developed by the NorESM team. We recently upgraded NorCPM to use the latest NorESM2-MM (Seland et al., 2020) that has contributed to CMIP6 to a range of MIPs (but not DCPP), which has a higher atmospheric resolution, notably improved overall biases and an improved marine biogeochemistry representation relative to NorESM1-ME (albeit at a tenfold computational cost).



Annan, J.D. (2005), Parameter estimation using chaotic time series. Tellus A, 57: 709-714. https://doi.org/10.1111/j.1600-0870.2005.00143.x



Athanasiadis, P.J., Yeager, S., Kwon, YO. et al. Decadal predictability of North Atlantic blocking and the NAO. npj Clim Atmos Sci 3, 20 (2020). https://doi.org/10.1038/s41612-020-0120-6



Bauer, P., Thorpe, A. & Brunet, G. The quiet revolution of numerical weather prediction. Nature 525, 47–55 (2015). https://doi.org/10.1038/nature14956



Chikamoto, Y., Timmermann, A., Widlansky, M. J., Zhang, S., and Balmaseda, M. A.: A Drift-Free Decadal Climate Prediction System for the Community Earth System Model, Journal of Climate, 32, 5967–5995, https://doi.org/10.1175/JCLI-D-18-0788.1, 2019.

Counillon, F., Keenlyside, N., Toniazzo, T. et al. Relating model bias and prediction skill in the equatorial Atlantic. Clim Dyn 56, 2617–2630 (2021). https://doi.org/10.1007/s00382-020-05605-8

Deser, C., Lehner, F., Rodgers, K.B. et al. Insights from Earth system model initial-condition large ensembles and future prospects. Nat. Clim. Chang. 10, 277–286 (2020). https://doi.org/10.1038/s41558-020-0731-2



Drews, A., Huo, W., Matthes, K., Kodera, K., and Kruschke, T.: The Sun's Role for Decadal Climate Predictability in the North Atlantic, Atmos. Chem. Phys. Discuss. [preprint], https://doi.org/10.5194/acp-2021-241, in review, 2021.



Klavans, J.M., Cane, M.A., Clement, A.C. et al. NAO predictability from external forcing in the late 20th century. npj Clim Atmos Sci 4, 22 (2021). https://doi.org/10.1038/s41612-021-00177-8



Liguori, G., McGregor, S., Arblaster, J.M. et al. A joint role for forced and internally-driven variability in the decadal modulation of global warming. Nat Commun 11, 3827 (2020). https://doi.org/10.1038/s41467-020-17683-7



Polkova, I.,  Brune, S.,  Kadow, C.,  Romanova, V.,  Gollan, G.,  Baehr, J., et al. (2019).  Initialization and ensemble generation for decadal climate predictions: A comparison of different methods. Journal of Advances in Modeling Earth Systems,  11,149–172. https://doi.org/10.1029/2018MS001439



Scaife, A.A., Smith, D. A signal-to-noise paradox in climate science. npj Clim Atmos Sci 1, 28 (2018). https://doi.org/10.1038/s41612-018-0038-4



Seland, Ø., et al.: Overview of the Norwegian Earth System Model (NorESM2) and key climate response of CMIP6 DECK, historical, and scenario simulations, Geosci. Model Dev., 13, 6165–6200, https://doi.org/10.5194/gmd-13-6165-2020, 2020.



Smith, D.M., Scaife, A.A., Eade, R. et al. North Atlantic climate far more predictable than models imply. Nature 583, 796–800 (2020). https://doi.org/10.1038/s41586-020-2525-0

Yeager, S. G., Danabasoglu, G., Rosenbloom, N. A. et al. (2018). Predicting Near-Term Changes in the Earth System: A Large Ensemble of Initialized Decadal Prediction Simulations Using the Community Earth System Model, Bulletin of the American Meteorological Society, 99(9), 1867-1886.



Comment 2: The authors rely on the ocean-atmosphere coupling to transfer observational information to the atmosphere. I am totally fine with this, especially in the view of multi-annual predictions. Nevertheless, what do the authors think about having an atmospheric assimilation as well, at least for the large scale atmospheric state? I would also like the authors to be very clear from the beginning that the EnKF is applied for oceanic/sea ice DA and IS NOT applied in atmospheric assimilation here. This point could be made in l.94 "NorCPM1 further stands out in that it uses an EnKF based anomaly DA scheme..."



We thank the reviewer for the comment. In the revised manuscript, we will be more clear from the beginning (l.94 and other places) that the EnKF updates are not applied to the atmosphere and land states. We will include a short discussion on the implications of currently not having atmospheric assimilation and future plans regarding adding it.



Utilizing atmospheric observations and better constraining the atmospheric circulation variability has potential to improve the ocean and sea ice initialisation by producing surface fluxes that are more consistent with the SST and SIC anomalies during the assimilation phase. Constraining the atmospheric circulation will also improve atmosphere and land initialisation, which would be beneficial for seasonal prediction. Not utilizing atmospheric observations, the ensemble spread of our prediction system is likely larger than the theoretical uncertainty of the ocean–sea ice initial state given all available observations. This is particularly true during the propagation phases—i.e., between the monthly assimilation updates—when the unsynchronised atmospheric forcing variability amplifies growth in the ensemble spread. While the Bjerknes feedback at least partly synchronises tropical atmospheric variability, the extratropical atmospheric variability of the individual simulation members remains largely unconstrained in NorCPM. As a result, the EnKF assimilation of the ocean and sea ice state has to work against the simulated intrinsic atmospheric variability that is not in phase with the observations. The success of the FOSI approach (Yeager et al., 2018) demonstrates the potential in solely constraining surface fluxes over ocean and sea ice for initializing multi-year climate predictions, indicating that synchronising the ocean circulation through surface fluxes of heat, freshwater and momentum can largely compensate for not utilizing subsurface observations. We expect that the combination of constraining atmospheric variability and performing ocean–sea ice assimilation will provide the best result for climate prediction, as demonstrated by Polkova et al. (2019).



Assimilation of atmospheric observations into NorCPM is work in progress. One challenge is to avoid a collapse of spread in the surface ocean that is needed for determining the ensemble covariance in EnKF assimilation. Another challenge is that atmospheric updates have to be performed at daily or higher frequency. NorCPM's monthly EnKF assimilation updates are currently performed offline (the model integrates for one month, writes restart conditions that are updated by the EnKF, reads the updated conditions and integrates the next month). An offline approach with high-frequent updates (e.g., Karspeck et al., 2018) would result in a computational overhead that we consider unacceptable and to perform EnKF-based atmospheric data assimilation we would need to move the assimilation step online (Zang et al. , 2007; Nerger et al., 2020). As a readily available alternative, we are exploring atmospheric nudging in combination with EnKF-based ocean–sea ice assimilation, a strategy that has been successfully tested in the MPI MiKlip system (Polkova et al., 2019). We will take advantage of the availability of multiple simulation members of the reanalysis products like ERA5 (Hersbach et al., 2020) and CERA (Laloyaux et al., 2018) and nudge the members of the NorCPM analysis to individual members of the reanalysis products. This will provide a representation of atmospheric observational uncertainties and help generate ensemble spread in the ocean state. We are aiming at complementing this approach with the leading average cross covariance technique (Lu et al., 2015) that can perform a one-way strongly coupled data assimilation (from atmosphere to ocean) and has been shown to allow improved ocean initialization taking advantage of the abundant atmospheric observation data.



Davis, N. A., Davis, S. M., Portmann, R. W., Ray, E., Rosenlof, K. H., and Yu, P.: A comprehensive assessment of tropical stratospheric upwelling in the specified dynamics Community Earth System Model 1.2.2 – Whole Atmosphere Community Climate Model (CESM (WACCM)), Geosci. Model Dev., 13, 717–734, https://doi.org/10.5194/gmd-13-717-2020, 2020.  



Hersbach, H, Bell, B, Berrisford, P, et al. The ERA5 global reanalysis. Q J R Meteorol Soc.  2020; 146: 1999–2049. https://doi.org/10.1002/qj.3803



Karspeck, AR, Danabasoglu G, Anderson J, et al. A global coupled ensemble data assimilation system using the Community Earth System Model and the Data Assimilation Research Testbed. Q J R Meteorol Soc.  2018; 144: 2404–2430. https://doi.org/10.1002/qj.3308     



Laloyaux, P., de Boisseson, E., Balmaseda, M., Bidlot, J.-R., Broennimann, S., Buizza, R., et al. (2018). CERA-20C: A coupled reanalysis of the twentieth century. Journal of Advances in Modeling Earth Systems,  10,  1172–1195. https://doi.org/10.1029/2018MS001273



Lu, F., Liu, Z., Zhang, S., & Liu, Y. (2015). Strongly Coupled Data Assimilation Using Leading Averaged Coupled Covariance (LACC). Part I: Simple Model Study, Monthly Weather Review, 143(9), 3823-3837. https://doi.org/10.1002/2015GL067238



Nerger, L., Tang, Q., and Mu, L.: Efficient ensemble data assimilation for coupled models with the Parallel Data Assimilation Framework: example of AWI-CM (AWI-CM-PDAF 1.0), Geosci. Model Dev., 13, 4305–4321, https://doi.org/10.5194/gmd-13-4305-2020, 2020.



Polkova, I., Brune, S., Kadow, C., Romanova, V., Gollan, G., Baehr, J., et al. (2019). Initialization and ensemble generation for decadal climate predictions: A comparison of different methods. Journal of Advances in Modeling Earth Systems, 11, 149–172. https://doi.org/10.1029/2018MS001439



Sun, J., Liu, Z., Lu, F., Zhang, W., & Zhang, S. (2020). Strongly Coupled Data Assimilation Using Leading Averaged Coupled Covariance (LACC). Part III: Assimilation of Real World Reanalysis, Monthly Weather Review, 148(6), 2351-2364.



Zhang, S., Harrison, M. J., Rosati, A., & Wittenberg, A. (2007). System Design and Evaluation of Coupled Ensemble Data Assimilation for Global Oceanic Climate Studies, Monthly Weather Review, 135(10), 3541-3564. https://doi.org/10.1175/MWR3466.1



Comment 3: The authors include a lot of figures, and I like this very much. However, the quality of the figure annotations (labels) is sometimes rather poor. I would like to ask the authors to re-assess the annotations, this would greatly help the reader to quickly connect with the figures.



Thanks for the suggestion. We will improve the quality of the figure annotations for the revised manuscript version (see examples in the Supplement to the reply). Also, we will move the Appendix C (baseline evaluation) and possibly Appendix D to a separate Supplementary Information document. This will reduce the number of figures in the main manuscript and allow us to reduce the compression of the figures.



Comment  3a: figures with maps are at the limit in terms of crowded information, but that is still okay.



If still acceptable in terms of crowded information then we prefer to keep the panel layout of the figures as is. The multi-panel layout allows the reader to visually compare the results for different lead years, prediction benchmarks and fields of interest.



Where possible, we tried to use a similar style for the figures with maps, mostly adopted from Yeager et al. (2018). The reader may need to spend some time to understand the first of such figures, but it should require less time to understand successive figures of the same style.



Comment 3b: although the maps themselves are in hires, their annotations sometimes look very lowres, e.g. as in Fig. 3



The poor label quality was a result of an unfortunate choice of font type, file format conversion and compression. As mentioned in the reply to 3, we will address the quality issue in the revised manuscript version (see example for new Figure 3 in the Supplement to the reply). 



Comment 3c: please put the annotations outside the maps according to rows and columns, similar to Fig.3 (Figs. 7,9-12,15-17,21-24,D3-5)



We will follow this suggestion in the revised manuscript (see example for new Figure 12 in the Supplementary Information to the reply).



Comment  3d: huge difference in font size, Fig. 14



The font size will be reduced in the revised manuscript (see example for new Figure 14 in the Supplement to the reply).



Comment 3e: Would be good to have all the a,b,c... at a rather similar position throughout all figures. Now they are sometimes in the upper left, lower left, or somewhere within the figure.



Good point. We will try to move the a,b,c... labels to the lower left corner of the panels for all figures of the revised manuscript.



l.98 "see Section 2.1.1 for details" Is this reference meant for the description of the DA? Then section 2.2 (or subsections) would fit better.



Thanks for noticing this. We will correct it to "see Section 2.2.3 for details".

We thank the reviewer for positive and constructive feedback. Below, we will address the comments one by one. 

I only have one suggestion to the authors that warrants mentioning outside the “specific comments” section. The comparison of skill between the hindcasts and/or assimilation simulations and the historical runs to assess the contribution of the forced response to skill currently relies on skill differences (i.e. ACC(hind)-ACC(hist) and so on). Recently, Smith et al. (2019) demonstrated that this practice does not fully capture the benefit of initialisation due to non-linear interactions in the system, and propose the use of residual ACC, that subtracts the forced signal (hist) from the hindcast signal prior to skill calculation. In their work, Smith et al. show that the use of residuals more accurately describes what the authors set out to assess here. I suggest the authors either present some results using residuals to illustrate this alternative metric, or at least discuss the residual as an alternative (and potentially superior) approach.

Thanks for the information and suggestions. 

We used ACC differences for mainly two reasons: to be able to directly compare with similar previous studies (e.g., Yeager et al., 2018), and to be able to evaluate skill benefits consistently with respect to different benchmarks (analysis, persistence, free historical). The use of ACC differences further allowed us to adopt the local and field significance testing from Yeager et al. (2018), whereas we would not know how to appropriately deal with uncertainty related to regressing out the forced signal. For these reasons we prefer not to replace our ACC differences plots. Instead we prefer to opt for presenting additional figures using the Smith et al. method in the Supplementary Information and briefly mention differences (if any) in the main manuscript.  We prefer not to present any results with the new method in the main manuscript, as we are limited concerning the number of figures we can add. We will add discussion on using the Smith et al. method (i.e., regressing out the forced model signal prior to skill calculation) as an alternative, more robust approach to computing ACC differences.    

Yeager, S. G., Danabasoglu, G., Rosenbloom, N. A., Strand, W., Bates, S. C., Meehl, G. A., Karspeck, A. R., Lindsay, K., Long, M. C., Teng, H., & Lovenduski, N. S. (2018). Predicting Near-Term Changes in the Earth System: A Large Ensemble of Initialized Decadal Prediction Simulations Using the Community Earth System Model, Bulletin of the American Meteorological Society, 99(9), 1867-1886.

l. 24-25 I do not think it is clear at this point what “non-assimilation experiments” are. Maybe just call them “NorESM1 simulations" for clarity?

Thanks for the suggestion. We will rephrase "non-assimilation experiments" to "NorESM1 simulations.

ll. 54-55 This statement could benefit from one or two publications to back up the claim.

Thanks for the suggestion. We will add the below citations.

Årthun, M., E. W. Kolstad, T. Eldevik, and N. S. Keenlyside (2018), Time Scales and Sources of European Temperature Variability, Geophys Res Lett, 45 (0), doi:10.1002/2018GL077401.

Athanasiadis, P. J., S. Yeager, Y.-O. Kwon, A. Bellucci, D. W. Smith, and S. Tibaldi (2020), Decadal predictability of North Atlantic blocking and the NAO, npj Climate and Atmospheric Science, 3 (1), 20, 10.1038/s41612-020-0120-6.

Sutton, R. T., , and D. L. R. Hodson, 2005: Atlantic Ocean forcing of North American and European summer climate. Science, 309 , 115–118.

Omrani, N. E., N. S. Keenlyside, J. Bader, and E. Manzini (2014), Stratosphere key for wintertime atmospheric response to warm Atlantic decadal conditions, Climate Dynamics, 42 (3-4), 649-663, 10.1007/s00382-013-1860-3.

ll. 63-74 Similarly, this entire paragraph can in my opinion not go without citations. Please add some published work to back up the statements made here.

Thanks for the suggestion. We will add citations as indicated below.

Current climate prediction systems are thought to not fully realise the predictive potential on multi-year times scales, although the practical limits of predictability themselves and their regional variations are poorly known (Branstator et al., 2012; Sanchez-Gomez et al., 2015; Smith et al., 2020).

The skill of climate prediction depends on the initialisation of internal climate variability state, the representation of the dynamics and processes that lead to predictability, and the representation of the climate responses to external forcings (Branstator and Teng, 2010; Latif and Keenlyside, 2011; Bellucci et al., 2015; Yeager and Robson, 2017).

Dynamical climate prediction systems typically use Earth system models (initially developed to provide uninitialised long-term climate projections) for representing the dynamics and the responses to external forcings (Meehl et al., 2009; Meehl et al., 2013).

Importantly, the dynamical prediction systems add initialisation capability to the ESMs, adopting a wide range of initialisation strategies (see Section 2.2.1) (Meehl et al., 2021).

A better understanding of the three aspects – initialisation, model dynamics, forcing response – is fundamental for better exploiting the climate predictive potential and improving estimates of climate predictability (Keenlyside and Ba, 2010; Cassou et al., 2018; Verfaillie et al., 2021).

The existing climate prediction systems undersample effects of model and initialisation uncertainty and are not necessarily well suited to address questions related to changes in the observing system. The benefits from using advanced data assimilation for initialisation, especially in an ocean density coordinate framework, are not well explored.

Bellucci, A., et al. (2015), Advancements in decadal climate predictability: The role of nonoceanic drivers, Rev Geophys, 53, 165–202, 10.1002/2014RG000473.

Branstator, G., and H. Y. Teng (2010), Two Limits of Initial-Value Decadal Predictability in a CGCM, J Climate, 23(23), 6292-6311

Branstator, G., H. Y. Teng, G. A. Meehl, M. Kimoto, J. R. Knight, M. Latif, and A. Rosati (2012), Systematic Estimates of Initial-Value Decadal Predictability for Six AOGCMs, J Climate, 25(6), 1827-1846

Cassou, C., Y. Kushnir, E. Hawkins, A. Pirani, F. Kucharski, I.-S. Kang, and N. Caltabiano (2018), Decadal Climate Variability and Predictability: Challenges and Opportunities, B Am Meteorol Soc, 99(3), 479-490, 10.1175/BAMS-D-16-0286.1.

Keenlyside, N. S., and J. Ba (2010), Prospects for decadal climate prediction, Wiley Interdisciplinary Reviews: Climate Change, 1(5), 627-635, 10.1002/wcc.69.

Latif, M., and N. S. Keenlyside (2011), A perspective on decadal climate variability and predictability, Deep Sea Research Part II: Topical Studies in Oceanography, 58, 1880-1894

Meehl, G. A., et al. (2009), Decadal Prediction Can It Be Skillful?, B Am Meteorol Soc, 90(10), 1467-1486, https://doi.org/10.1175/2009BAMS2778.1.

Meehl, G. A., et al. (2013), Decadal Climate Prediction: An Update from the Trenches, B Am Meteorol Soc, 10.1175/bams-d-12-00241.1.

Meehl, G. A., et al. (2021), Initialized Earth System prediction from subseasonal to decadal timescales, Nature Reviews Earth & Environment, 2(5), 340-357, 10.1038/s43017-021-00155-x.

Sanchez-Gomez, E., C. Cassou, Y. Ruprich-Robert, E. Fernandez, and L. Terray (2015), Drift dynamics in a coupled model initialized for decadal forecasts, Climate Dynamics, 1-22, 10.1007/s00382-015-2678-y.

Smith, D. M., et al. (2020), North Atlantic climate far more predictable than models imply, Nature, 583(7818), 796-800, 10.1038/s41586-020-2525-0.

Verfaillie, D., F. J. Doblas-Reyes, M. G. Donat, N. Pérez-Zanón, B. Solaraju-Murali, V. Torralba, and S. Wild (2021), How Reliable Are Decadal Climate Predictions of Near-Surface Air Temperature?, J Climate, 34(2), 697-713, 10.1175/JCLI-D-20-0138.1.

Yeager, S. G., and J. I. Robson (2017), Recent Progress in Understanding and Predicting Atlantic Decadal Climate Variability, Current Climate Change Reports, 3(2), 112-127, 10.1007/s40641-017-0064-z.

l. 80-81 Take it or leave it: to me, the order historical->assimilation->hindcasts would make more sense.

Thanks for the suggestion. We will rephrase it to "two sets of DCPP coupled reanalysis simulations, and two sets of initialised DCPP hindcast simulations that obtain their initial conditions from the two reanalysis sets.". 

l. 131, 134 Naïve question: is NorCPM1 a purely “physical model”?

Because we realized that the original formulation "physical model" was inaccurate, we will rephrase the formulation to "The Earth system model used in NorCPM1". 

NorESM1 does feature ocean biogeochemistry (which albeit does not influence the physical climate) as well as atmospheric chemistry. Also, some parameterizations of NorESM1 are based on empirical statistical relations rather than on first physical principles. Nevertheless they are intended to model physical processes. The EnKF does exploit statistical relations between physical variables, but relies on the ESM to obtain these relations.   

l. 302-304 Why the two different baseline periods? This might be obvious, but I am struggling to understand the motivation for this choice.

We used two periods primarily because OISST data were not available prior September 1981. We will add this information and modify the sentence to "but over the period 1982-2010 when assimilating OISSTV2 data (i.e., beyond 2010) because OISSTV2 was not available before this period."

In hindsight, it would have been more consistent if we had extended the OISST back in time with HadISST2 data (like we did for assim-i2) and then used the period 1980-2010. But we checked and found the 1982-2010 OISST-based climatology and the 1980-2010 HadISST2/OISST-based climatology (1980-1981 HadISST, 1982-2010 OISST) are practically indistinguishable. 

l. 316 I suppose “The approach” here references the SCDA approach used in assim -i2?

This is correct. 

In RC1, the reviewer noted that characterising assim-i2 as SCDA can be misleading because the term is typically used with respect to ocean­ and atmosphere and involves either atmospheric state update or use of atmospheric observations. 

We will therefore rephrase the text in question to "In assim-i2, we allow the oceanic observations to update the ocean and the sea ice components. In this case the system is a strongly coupled DA system (SCDA) with respect to the ocean­ and sea ice components, where the oceanic observations influence the sea ice component of the system both at the DA step and during the model integration. To avoid confusion with atmosphere-ocean SCDA (e.g., Penny et al., 2019), we will refer to the assim-i2 system approach as SCDA-OSI. The SCDA-OSI assures a more consistent initialisation across the ocean and sea ice components and exploits the longer temporal coverage of oceanic observations relative to sea ice observations (see also Appendix A)." 

Penny, S. G., Bach, E., Bhargava, K., Chang, C.-C., Da, C., Sun, L., and Yoshida, T.: Strongly Coupled Data Assimilation in Multiscale Media: Experiments Using a Quasi-Geostrophic Coupled Model, Journal of Advances in Modeling Earth Systems, 11, 1803–1829, https://doi.org/10.1029/2019MS001652, https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/2019MS001652, 2019.

ll. 414-416 ACC is known to be sensitive to spurious trends as a skill metric (e.g. Smith et al., 2019). RMSE-based metrics such as MSSS might be more realistic when it comes to skill assessment. While I see why the authors choose ACC for this study, I think a short sentence on potential ACC shortcomings might be in order here or in the discussion.

We will add some text on potential ACC shortcomings related to spurious skill and amplitude errors based on the reasoning below. We will also consider adding a MSSS map figure to the Supplementary Information of the manuscript to illustrate overall similarity between the ACC and MSSS measures (see Supplement to this reply).   

While we would reserve the term "spurious trends" for trends that e.g. arise from observational errors (probably not what the reviewer is hinting at), we generally agree with the reviewer's statement. ACC scores of multi-year averages are subject to spurious correlations because the climate signal of the real world can covary with the numerical predictions just by chance during the limited evaluation window (which can be interpreted as sampling error due to undersampling of climate variability). In case of such random-covariability, the amplitudes of the signals tend to differ since there is no apparent reason they should be the same. The MSSS would panelize for the amplitude error while the ACC does not, making the MSSS more robust to effects of spurious correlations. Our interpretation of Smith et al. (2019) is, however, that they consider MSSS as overly pessimistic because of penalization of amplitude errors that potentially can be corrected posteriori and they therefore favor the use of ACC. 

Initially, we prepared Figures for both ACC and MSSS. While the MSSS scores results were generally lower than the ACC scores (because of penalization of amplitude errors in the former), they were qualitatively consistent with the ACC scores and did not change our general skill assessment. We dropped the MSSS because the key literature we compare to favors ACC and also we realized that presenting both ACC and MSSS and discussing subtle differences between them would make the paper too long. 

Figure 2 Are these global metrics?

The metrics shown in Figure 2 are global, considering both vertical and horizontal dimensions. They are computed as detailed in Equations 2–7 in Section 3.1.1.

We will change the Figure caption to "Global assimilation statistics (see Section 3.1.1. for definitions)." to make this clearer. We will also correct the numbering of the Equations from 2–7 to 1–6.  

l. 465 There is a “.” Missing after “observations”.

Thanks. We will fix it in the revised version.

l. 507 Is this referencing “changes” due to data assimilation? I suggest being explicit about that.

Thanks for the suggestion. We will make clear this paragraph continues to discuss the impact of the assimilation on the mean state and will rephrase the text to "Some impact of DA on the mean state of assim-i1 is also seen...". 

ll. 532-533 I am struggling to see the smaller improvement for S300 compared to T300 reflected in figure 7. I suppose the authors reference figures 7f and 7g here, comparing the assimilation simulations’ improvement relative to historical simulations for T300 and S300? When I visually compare 7f to 7g, I am struggling to make out a clear “winner”. If anything, it appears to me that S300 shows slightly higher values. Could you please comment on that? In general, the text could at times benefit from more clear reference of figure sub-panels when making statements in the text.

Thanks for noting this. We will remove the sentence "The improvements for S300 are smaller ...". We made a mistake in our first version of Figure 7, resulting in smaller ACCs for S300. Later, we noticed and rectified this but forgot to correct the manuscript text. 

Also, we will try to refer more explicitly to sub-panels in the revised version of the manuscript.

Figure 7 The labelling of assim-i1 and assim-i2 as ANA1 and ANA2 in the figures is not optimal, as it is inconsistent with the labelling in the text. I suggest changing the labels in the figures for consistency. This is an issue in all figures that show skill comparison including assimilation simulations.

Thanks for the suggestion. We will make the labels/acronyms used on the figures consistent with those used in the manuscript text (we will also move the labels out of the panels as suggested by RC1). We will have to check whether the figure labels are still readable or become too small when changing to the long acronyms. Otherwise, we will use short acronyms also in the manuscript text. In addition, we will add a table that summarizes essential information of the different experiments. 

l. 559 ff. This discussion is reminiscent of work done by Koul et al. (2020) on which SPG index represents which underlying physical processes. I think this part could be shortened by referring to the above mentioned paper.

The text in question was not intended as discussion but rather to provide some minimum motivation and background for the readers less familiar with the North Atlantic subpolar gyre concept. We therefore prefer to keep the text but will add the citation to Koul et al. (2020) for further reading. 

l. 608-610 I think it is important to point out this “small contribution” is insignificant (is it?).

Thanks for pointing this out. 

We have computed a p-value of 0.085 using block-bootstrapping—with block lengths of 5 years—and re-sampling of members in the computation of the ensemble mean (see Appendix B for details on p-value computation).

Following Yeager et al. (2018), we have used a 10 % significance level throughout the manuscript and it would be inconsistent if we switched to 5 % significance level for this test. Results presented in Appendix D in Figure D2 are also suggestive for a small, but systematic influence from external forcing on simulated NINO34 SSTs, with positive correlation values for all calendar months.  

We will moderate the statement to "The ensemble mean of historical has a smaller amplitude and is only marginally correlated with the observed index (r=0.2, p=0.085, alpha=0.1), suggesting a potential small contribution from external forcing.".

l. 676-678 At least one citation should be given for the statement on internal vs external causes of the global warming hiatus (e.g. Medhaug et al. 2017?).

Thanks for the suggestion. We will cite Medhaug et al. (2017). 

l. 696-698 I assume the authors used ACC**2 to calculate explained variance? This should be made explicit.

Thanks for the suggestion. We will rephrase the sentence to "One should note that a change in correlation from 0.6 to 0.9 equates to more than doubling in explained variance from 36 % to 81 % (estimated by the square of the correlation)." 

ll. 741-754 In a recent paper, Borchert et al. (2021) showed increased contribution of forcing to decadal SPNA SST prediction skill in CMIP6 compared to CMIP5, using a multi-model ensemble including NorCPM1. How do the authors square their findings presented here with what Borchert et al. (2021) found?

We cannot say with certainty how our findings square with Borchert et al. (2021), who found an increasing contribution from external forcings to SPNA skill in CMIP6 relative to CMIP5. It would be interesting to know the isolated effect of including NorCPM1 CMIP6 DCPP output in their multi-model analysis. It is possible that NorCPM1 represents a model outlier. 

Our Figure 13c shows that the EN4-based observational estimate of upper SPNA temperature is negatively correlated with the ensemble mean of NorCPM1's historical experiment over the period 1950-present. This is contrasting the general positive SPNA skill contribution from external forcing that Borchert et al. found in multi-model ensembles (e.g., their Fig. 1). While less clear than for upper ocean temperature, SST is also negatively correlated over SPNA (Figure 14c) with a correlation value below -0.4 (Figure 14e). Moreover, Figure 12 suggests that the ACCs for SST in the SPNA region of the historical experiment are predominantly negative because (ACC_hindcast-i1 - ACC_historical) is larger than ACC_hindcast-i1 in most of the region. Notwithstanding the above, it is possible (but we have not checked it) that NorESM1's historical experiment using CMIP5 forcings features even more negative ACCs for SST in the SPNA region than NorCPM1's historical experiment with CMIP6 forcings, somewhat consistent with Borchert et al. (2021). 

In the revised manuscript, we will point out the apparent discrepancy to the findings of Borchert et al. (2021). Following the sentence with "poorly performing simulated forced trend" (line 743) we will add "This result stands in contrast to multi-model findings (that include NorCPM1) suggesting a positive contribution of the forced signal to SPNA temperature skill over a comparable period (Borchert et al., 2021)."  

As detailed in Appendix C (lines 1311-1325), we suspect a problem with CMIP6 land use change specification, leading to an unrealistic historical cooling trend over North America in NorCPM1. Via downstream effects, the continental cooling (which we suspect is an artifact) may contribute to the SPNA cooling trend shown after 1980 (e.g., Fig. 13c), exacerbating the discrepancy between the observed and simulated SPNA temperature evolutions. How rectifying NorCPM1's CMIP6 land use change specification and also how replacing NorESM1 with NorESM2 may impact SPNA prediction skill (both initialised and uninitialised) in NorCPM is subject to future investigation.

l. 825 To me, the phrase “potential predictability” refers to the skill of simulations initialised from a piControl simulation with respect to the same piControl simulation. What the authors demonstrate here (skill of hindcasts initialised from a reanalysis-type simulation with respect to a reanalysis that did not directly assimilate biogeochemistry, but produces biogeochemistry that is consistent with observed physical climate) is to me more than that. The authors might want to re-consider their phrasing here so as to avoid underselling their findings.

As the manuscript text states, we adopted the definition of "potential predictability" from Yeager et al. (2018), a key reference for our study.

We agree that our use of "potential predictability" differs from the one in some earlier studies that utilized a piControl simulation (e.g., Collins et al., 2006) and we will more clearly point this out in the revised version. 

We will rephrase the text to "Following Yeager et al. (2018), we therefore also analysed the model's ability to hindcast its own analysis over the period 1960–2018 (Fig. 16). We will refer to this as the potential* predictability, using the asterisk to indicate that it differs from more conventional potential predictability estimates based on self-prediction that typically utilize a pre-industrial control simulation (e.g., Collins et al., 2006)."

Collins, M., Botzet, M., Carril, A. F., Drange, H., Jouzeau, A., Latif, M., Masina, S., Otteraa, O. H., Pohlmann, H., Sorteberg, A., Sutton, R., & Terray, L. (2006). Interannual to Decadal Climate Predictability in the North Atlantic: A Multimodel-Ensemble Study, Journal of Climate, 19(7), 1195-1203.

l 888 It should be “in the Pacific sector”.

Thanks. We will correct this. 

Figs. 25 & 26 The comparison of mean skill and skill of the mean signal is interesting. However, the authors only mention Fig. 26 briefly in one paragraph. Would 1-2 more sentences on this topic be interesting to a wider readership?

Thanks for the comment. We will try to add a few sentences for the wider readership based on the below. Please let us know if you had something different in mind. 

We consider the mean skill of local (grid-cell scale) sea ice concentration as a simple integrated metric for the sea ice performance of our prediction system that complements the skill of the hemispheric mean. It illustrates, for example, that skill reemergence in the second year seen in the total sea ice area does not necessarily translate into statistically significant local prediction skill during the second year. Presumably, this is partly related to the lack of skill for the Arctic oscillation which primarily causes a regional re-distribution of sea ice with limited effect on the hemispheric mean, whereas the hemispheric mean is more controlled by the externally forced trend. Efforts that target sea ice prediction from a climate service perspective are advised to do more elaborated analysis (that were beyond the scope of our paper), like predicting the temporal evolutions of principal components or regional averages of sea ice concentration, to additionally consider the predictability of atmosphere and subsurface ocean variability, and to perform statistical/dynamical downscaling.  

l 1253 ff I particularly like this (brief) mention of observational error, particularly in light of the small but important differences between assim-i1 and assim-i2. This might be a candidate for the main text, as this info is currently a little hidden.

Thanks for the suggestion. We will move this paragraph to the main manuscript. 

l 1281 As far as I can make out, the mean state of AMOC is not shown in Figure C3c, nor is the vigorous nature of simulated AMOC. This might be a phrasing issue, but I was looking for this information in the figure. Rephrase to avoid this in the future?

We are a bit unsure about this comment. Could the reviewer elaborate?

Figure C3c shows the climatological Atlantic meridional overturning circulation (AMOC) streamfunction in units Sv for NorCPM1. To our knowledge, this is the standard way for characterising the mean state of AMOC. Values in excess of 30 Sv clearly indicate the vigorous nature of the simulated AMOC ("vigorous" in the sense of strong, not necessarily variable). While this is somewhat different from the "time-mean AMOC strength evaluated at a fixed latitude", these two characterizations are closely related. We will rephrase the figure caption text from "Atlantic meridional overturning streamfunction" to "Atlantic meridional overturning circulation (AMOC) streamfunction" to be more precise.

The overly strong AMOC has been a longstanding, well-documented problem of previous NorESM versions (e.g., Bentsen et al., 2013, Cheng et al., 2013) but has been mitigated through updates and re-tuning of the ocean code in more recent NorESM versions (Guo et al., 2019; Seland et al., 2020). 

Bentsen, M., Bethke, I., Debernard, J. B., Iversen, T., Kirkevåg, A., Seland, Ø., Drange, H., Roelandt, C., Seierstad, I. A., Hoose, C., and Kristjánsson, J. E.: The Norwegian Earth System Model, NorESM1-M – Part 1: Description and basic evaluation of the physical climate, Geosci. Model Dev., 6, 687-720, https://doi.org/10.5194/gmd-6-687-2013, 2013.

Cheng, W., Chiang, J. C. H., & Zhang, D. (2013). Atlantic Meridional Overturning Circulation (AMOC) in CMIP5 Models: RCP and Historical Simulations, Journal of Climate, 26(18), 7187-7197.

Guo, C., Bentsen, M., Bethke, I., Ilicak, M., Tjiputra, J., Toniazzo, T., Schwinger, J., and Otterå, O. H.: Description and evaluation of NorESM1-F: a fast version of the Norwegian Earth System Model (NorESM), Geosci. Model Dev., 12, 343–362, https://doi.org/10.5194/gmd-12-343-2019, 2019.

Seland, Ø., Bentsen, M., Olivié, D., Toniazzo, T., Gjermundsen, A., Graff, L. S., Debernard, J. B., Gupta, A. K., He, Y.-C., Kirkevåg, A., Schwinger, J., Tjiputra, J., Aas, K. S., Bethke, I., Fan, Y., Griesfeller, J., Grini, A., Guo, C., Ilicak, M., Karset, I. H. H., Landgren, O., Liakka, J., Moseid, K. O., Nummelin, A., Spensberger, C., Tang, H., Zhang, Z., Heinze, C., Iversen, T., and Schulz, M.: Overview of the Norwegian Earth System Model (NorESM2) and key climate response of CMIP6 DECK, historical, and scenario simulations, Geosci. Model Dev., 13, 6165–6200, https://doi.org/10.5194/gmd-13-6165-2020, 2020.